linear optocoupler, what accuracy can I expect?

[wanna_eat_1kg_of_elecons]

Hello all,
I am looking at galvanic isolation for analog signals. I think I need a linear opto as there is a feedbackloop which will compensate for temperature and aging of the part. I am new to this subject and read some datasheets,

like these ones:
https://www.ixysic.com/home/pdfs.nsf/www/LOC110.pdf/$file/LOC110.pdf
https://www.ixysic.com/home/pdfs.nsf/www/AN-107.pdf/$file/AN-107.pdf

(1)The main question is what accuracy can I expect? In the AN-107 they tell the accuracy of the LOC110 is max 14 bits is this just advertisement talk or can I really achieve this?

(2) Of course there are other brands selling similar devices. Like vishay sells the IL300 etc. In those documents they talk about maximum 12 bits accuracy. I have not found devices with a higher accuracy (higher than 14 bits). But maybe there are devices that can easy do higher accuracy, has anyone got experience with these devices?

(3)I ask this because I want to use them and learn how to calculate accuracies It would really like to get a idea what they are talking about for instance in the application note of the IL300 datasheet they tell:

"
The IL300’s 130 dB common mode
rejection (CMR), ± 50 ppm/°C stability, and ± 0.01 %
linearity provide a quality link from the sensor to the
controller input.
"

how do I interpret this? the application note can be found here:
**broken link removed**

the text is on page page 1.

Hope someone of you can learn me how to use these devices! Thank you for the help!

 

[MrAl]

Hi,

How much accuracy do you really need? That's the first question to answer.

Most part specs are pretty good, but then there are the real world factors that come into play which makes them look less attractive. So the rule then would be to get something that is better than what you need so you want be sorry. Also, for a critical application it would not be uncommon for a manufacturer to buy several different parts and do their own in house testing, which would tell you more than the data sheet. So dont be afraid to try a few different parts and see what is going to work in your application. It is also good to get the parts from a few different sources if the application is going into production so you can get a feel for the specification spread over different part lots. When the parts are tested the board should be as close as possible to the real thing, the end product board and whatever else is required.

Another more reliable way to obtain galvanic isolation is to go completely digital. This method is not dependent on any linear part specification because the data is transferred digitally so a single regular type opto coupler can be used. The idea is to encode the data using an A to D converter and then send the digitized data to the main circuit through an opto coupler. The main circuit then interprets the digital data.
This requires a small isolated power supply to power the front end ADC, so it may not be applicable for every application unless an isolated power supply can be added. The benefits however are that you skip the linear opto isolator requirement and should get long life out of it without too much worry.

 

[Nigel Goodwin]

I agree with MrAL, do the conversion BEFORE the isolation, and then transfer simple serial data via a cheap opto-coupler.

 

[crutschow]

What is the analog signal that you need to isolate?

 

[wanna_eat_1kg_of_elecons]

Hi,

//(A)
How much accuracy do you really need? That's the first question to answer.

//(B)
Most part specs are pretty good, but then there are the real world factors that come into play which makes them look less attractive. So the rule then would be to get something that is better than what you need so you want be sorry. Also, for a critical application it would not be uncommon for a manufacturer to buy several different parts and do their own in house testing, which would tell you more than the data sheet. So dont be afraid to try a few different parts and see what is going to work in your application. It is also good to get the parts from a few different sources if the application is going into production so you can get a feel for the specification spread over different part lots. When the parts are tested the board should be as close as possible to the real thing, the end product board and whatever else is required.

//(C)
Another more reliable way to obtain galvanic isolation is to go completely digital. This method is not dependent on any linear part specification because the data is transferred digitally so a single regular type opto coupler can be used. The idea is to encode the data using an A to D converter and then send the digitized data to the main circuit through an opto coupler. The main circuit then interprets the digital data.
This requires a small isolated power supply to power the front end ADC, so it may not be applicable for every application unless an isolated power supply can be added. The benefits however are that you skip the linear opto isolator requirement and should get long life out of it without too much worry.

MrAl,

Thank you for your reply!!

I did put comment sings in "//" to make it easier for me to reply

(A)What I try to do is make a design to learn things, basically it is a linear PSU with 2 transformers, and one micro-controller (uC). So the smart transformer part will have the uC. The other part with no uC needs information about the voltage set and the current set. I will try to set the voltage and current with 2 pwm outputs of the uC feed this signal trough 2 normal opto-coupler to the "stupid" part and read it back with 2 linear optocouplers to the smart uC part of the PSU. 2 as I want to read and set Voltage (thats one) and Current (thats2). As the pinout of the IL300 and LOC110 is the same I can order them both and test them both thank you for the advice! So to get back to your point (A) what resolution do I really need? I know there are 14 bits parts but I want to learn if there is a analog part that can give me a higher accuracy. If so I want to put it in a test pcb and test this option also.

(C) I know I could also make the signal digital first, it is probably easier to realize for me but that way I wont learn a lot of new things. For instance I do not know if it is possible to transfer a galvanic isolated analog 16 bits signal and I will never know if it is easier or harder if I do not try to get this knowledge.


thank you all for your replies so far!

 

[MrAl]

Hi,

Well ok then since this is a learning experiment and not a product research question, what you probably should do then is, *both*.

That is, create the digital channel first so you can make sensible measurements with reasonable accuracy. You can then buy various analog parts and see how good they compare to the digital channel. You can use a good DAC to drive the experiments. You can then answer this question first hand.

 

[crutschow]

Few applications require better than 0.1% accuracy, particularly in a power supply. Why do you think you need .0015%?